Know What's Happening in Business

Columbia CS Professor: Why LLMs Can’t Discover New Science

In this episode, Columbia computer‑science professor Vishal Misra explains why large language models, despite their rapid progress, are unlikely to generate truly novel scientific breakthroughs because they lack causal understanding and grounding in real‑world experimentation. He delves into why chain‑of‑thought prompting improves performance by breaking problems into intermediate steps, outlines what a genuine AGI would need beyond pattern‑matching, and identifies the root causes of LLM hallucinations as distributional gaps and over‑confidence in generated text. Misra’s perspective blends deep expertise in AI theory with practical insights on the current limits of LLMs.